Re: [PATCH] perf_event: fix error handling path

From: jovi zhang
Date: Thu Dec 09 2010 - 21:27:56 EST


On Fri, Dec 10, 2010 at 1:17 AM, Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> wrote:
> On Thu, 2010-12-09 at 09:06 -0800, Corey Ashford wrote:
>> On 12/07/2010 05:51 PM, jovi zhang wrote:
>> > On Mon, Dec 6, 2010 at 9:59 AM, jovi zhang<bookjovi@xxxxxxxxx> Âwrote:
>> >> On Sun, Dec 5, 2010 at 8:29 PM, Thiago Farina<tfransosi@xxxxxxxxx> Âwrote:
>> >>>
>> >>> On Sat, Dec 4, 2010 at 1:19 AM,<bookjovi@xxxxxxxxx> Âwrote:
>> >>>> fix error handling path
>> >>>>
>> >>>> Signed-off-by: Jovi Zhang<bookjovi@xxxxxxxxx>
>> >>>> Â kernel/perf_event.c | Â Â2 --
>> >>>> Â 1 files changed, 0 insertions(+), 2 deletions(-)
>> >>>>
>> >>>> diff --git a/kernel/perf_event.c b/kernel/perf_event.c
>> >>>> index cb6c0d2..62f9e9d 100644
>> >>>> --- a/kernel/perf_event.c
>> >>>> +++ b/kernel/perf_event.c
>> >>>> @@ -1918,8 +1918,6 @@ static int get_callchain_buffers(void)
>> >>>> Â Â Â Â }
>> >>>>
>> >>>> Â Â Â Â err = alloc_callchain_buffers();
>> >>>> - Â Â Â if (err)
>> >>>> - Â Â Â Â Â Â Â release_callchain_buffers();
>> >>>
>> >>> Care to explain in the change log message? As I reader, is not clear
>> >>> to me what is wrong with this.
>> >>
>> >> Sorry, the description should be as:
>> >> fix error handling path. alloc_callchain_buffers() can return -ENOMEM,
>> >> in this time callchain_cpus_entries maybe is NULL, It will oops if
>> >> invoke release_callchain_buffers() when callchain_cpus_entries is
>> >> NULL.
>> >>
>> > I hope my understanding is right, is it?
>>
>> One possible problem here is what if it returns an error other than
>> -ENOMEM, and the buffers do need to be released? ÂMaybe you could change
>> the code to
>>
>> err = alloc_callchain_buffers();
>> if (err != -ENOMEM)
>> Â Â Â release_callchain_buffers();
>>

Thanks for you suggestion. I also thought it, but I think it should
release the buffers in alloc_callchain_buffers if there have some
error(like many other part of kernel code).

It's not a good way to trace return error code outside of
alloc_callchain_buffer, if there have many return error code in
future, maybe need to write code like this:
if(err && err != -ENOMEM && err != -EXXX ) ...

It's very ugly. is right?

>>
>> Currently, alloc_callchain_buffers cannot return any error code other
>> than -ENOMEM, but that might change in the future.
>
> The kernel convention is to fully clean up after yourself if you return
> an error. So in that sense the patch seems right.
>
> Anyway, anybody care to post a new patch with slightly extended
> changelog?
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/